Upload
vancong
View
214
Download
0
Embed Size (px)
Citation preview
Towards Measuring Relational Embeddedness:
2 Factor Analyses of TRENDS Pilot Survey Data
What is …? Relational Embeddedness. A theoretical con-
struct that attempts to describe reasons why per-
sons maintain certain interpersonal relation-
ships.
The specific theory of interest was formulated
by Hite (2001). Simply stated relational em-
beddedness is a function of the level to which
an individual’s relationship involves more or
less of three components:
Dyadic Interaction: The extent and quality of
interpersonal interaction.
Personal Relationship: Amounts of the emo-
tional connections in the relationship.
Social Capital: The level of mutual and commu-
nal reciprocity affecting the relationship
What is…? Factor Analysis. A statistical method which uses
analysis of the ways in which survey or test
items tend to be answered in the same ways to
empirically estimate the degree to which items
are related to one another and to latent con-
structs.
TRENDS. A survey designed to measure the
levels of three theoretical constructs present in
relationships. Ultimately the survey is designed
to be used in network studies.
Network Studies: A genre of research which
analyzes relationship patterns. To use a survey
such as TRENDS in a network setting will re-
quire methods of assessing and controlling for
non-independence as participants will be asked
to complete surveys for as many relationships
as appropriate given their network standing and
the purpose of the instrument.
DI
TYP_2 e3
1
1
RIF_P5 e41
RIE_G1 e51
RIE_Q5 e61
RIQ_W5 e71
RIF_E2b e81
PR
TYP_3e9
1
1
RPS_K4e101
RPP_K3e111
RPA_L1e121
SC
TYP_1Be13
RSD_R3e14
RSD_E3e15
RSN_B3e16
1
1
1
1
1
RIX_F3
RIX_D2 e11
e21
FACTOR 2
RSNB1 e31
SWZZ99 e41
EMBE1 e51
RSDA5 e61
SWI5 e71
TP5 e81
FACTOR 2
SWE2e9
1
1
RPAL1e101
RIQS2e111
EMBV3e121
FACTOR 3
TS5e13
SWR3e14
RSNS4e15
EMBF1e16
1
1
1
1
1
RPSS2
RIFE4A1 e11
e21
FACTOR 1 1
Study 2: Background The data analyzed for this study was generated as part
of a larger survey conducted with school head teachers
in Uganda. The head teachers were asked to answer
questions about relationships with other head teachers
which provided them with resources beneficial to the
accomplishment of their work. This network of school
administrators was defined geographically by district (a
Ugandan political division, not equivalent to a US
school district). However, the respondents were not lim-
ited in choosing the relationships they rated to only
their relationships with other head teachers in the same
district.
This type of study design may be helpfully pictured
with a network diagram or map like the one below in
which individuals are displayed as circles and the rela-
tionships between them are line segments.
As part of this study a number of items which had been
included in the TRENDS II piloting were asked regard-
ing each relationship. Many of these items were elimi-
nated from the final TRENDS II factor models due to
factor loadings which did not correspond to the theory-
based latent constructs. The first step in this analysis
was to conduct exploratory factor analyses of these
items to determine an appropriate factor model which
could be tested in CFA using the M+ program.
Study 2: Hypotheses Having established an empirical factor model, the next
step was to test the model in a confirmatory analysis to
determine its model fit and factor loading characteris-
tics.
Steps 2a and b were to conduct two additional CFA’s in
which the identity of the survey respondent and the
identity of the survey’s “target” (subject) was used to
cluster the data in order to control for the effects of the
same person filling out multiple questionnaires or being
the “target” of multiple filled out questionnaires.
Specifically, the hypotheses were that:
The identified factor structure would be statistically sig-
nificant and exhibit fair model fit, factor loading and co-
variance statistics.
The effect of survey respondent would be significant,
leading to improvements in model fit over the 1st model
due to the clustering of the survey respondents.
The effect of survey subject would not make significant
improvement in model fit due to the diffuse nature of
the network subjects as illustrated in the above network
diagram.
Study 2: Results & Discussion Values closer to 1.0 are desired for CFI and TLI meas-
ures of model fit. RMSEA ideal values are closer to
zero.
The respondent-clustered CFA returns the best model fit
statistics. This is in keeping with the hypotheses.
As TRENDS moves towards use in the intended net-
work settings clustered CFA analyses will be crucial.
Factor Analysis Conducted by Tim Walker, PhD student in Educational In-
quiry, Measurement & Evaluation
In partial fulfillment of the course requirements of Sociology 706r, taught by
Dr. Joseph Olson
Social networks studies data collected by Dr. Julie Hite, Department of Edu-
cational Leadership & Foundations, McKay School of Education, BYU
Study 1: Background The two previous TRENDS pilot survey validation
studies had utilized the BYU teaching faculty as the
survey population. Each professor who agreed to par-
ticipate filled out the survey form regarding one of their
work-related relationships.
The data utilized in this study was similarly generated
by surveying a sample of the BYU faculty. Each profes-
sor who agreed to participate was asked to select a sin-
gle individual and answer 45 items dealing with their
work-related relationship with that individual. The par-
ticipants were asked to choose a person with whom
they have interacted, but who is not a member of their
own college.
Among the 45 items in the TRENDS III were several
items which had not been piloted as part of the
TRENDS I and II pilots. This was due to the poor per-
formance of some of the existing items in the TRENDS
II analyses, which included CFA of TRENDS items for
the first time.
The data were collected on paper copies of the survey.
Undergraduate research assistants (URA’s) contacted
the sampled faculty and made arrangements to invite
them to participate. The URA’s then arranged for the
retrieval of the surveys.
Study 1: Hypotheses Two essential hypotheses were tested in this study. First,
the hypothesized factor structure of the 45 items identi-
fied in the TRENDS II piloting were analyzed to deter-
mine if they still represented a model for the survey
which was statistically significant and returned good
model fit statistics.
The second hypothesis was that through an iterative
process like that outlined above a significant, well-
fitting model could be identified with a reduced number
of items.
The second hypothesis was crucial as it represents a vi-
tal step in the evolution of the TRENDS from the pi-
loted single-response studies to the desired network
study.
Study 1: Discussion The hypothesized TRENDS III factor structure, based
upon TRENDS II analysis and Hite’s relational em-
beddedness theory did not exhibit good model fit. This
was largely due to the performance of the newly piloted
items.
By eliminating these items from models used within the
iterative process illustrated above to identify a shortened
survey instrument, and by simplifying the factor struc-
ture to three factors, good model fit was obtained for the
shortened model. In short, while only batting .500 we
still managed to hit “at least a double, and perhaps a
homerun.” (Olsen 2010)
Future Studies 1. The sixteen item survey should be utilized
in a network setting. This would enable
analysis of the appropriateness of these spe-
cific items and factors in this type of study
setting. Additionally, it might allow work
on the following questions:
A. When respondents answer multiple
surveys what are appropriate
ways of measuring respondent
influence on the subsequent relation-
ship scores.
B. When individuals are the subject of
multiple surveys what are
appropriate ways of measuring sub-
ject influence on the subsequent
relationship scores.
C. At what level of clustering in respon-
dent and subject is statistical control
necessary?
An iterative
approach to re-
ducing instru-
ment size and
retaining fac-
tor structure
Theoretical pedi-
gree and validation
process of
TRENDS
TRENDS 16 item
Factor Structure TRENDS U
Factor Structure
Study 1: Results
Works Cited
Granovetter M. 1983. The strength of weak ties: A network theory re-
visited. Sociological Theory.1: 201-233.
Hite J. 2003. Patterns of multidimensionality among embedded net-
work ties: A typology of relational embeddedness in emerging en-
trepreneurial firms. Strategic Organization. 1:9-49.
Muthen & Muthen. 2008. M+ 5.2 (Statistical Analysis Software)
Olsen, J. 2010. Personal Communication.